Search Images Maps Play YouTube News Gmail Drive More »
Sign in
Screen reader users: click this link for accessible mode. Accessible mode has the same essential features but works better with your reader.

Patents

  1. Advanced Patent Search
Publication numberUS4991087 A
Publication typeGrant
Application numberUS 07/233,601
Publication dateFeb 5, 1991
Filing dateAug 18, 1988
Priority dateAug 19, 1987
Fee statusLapsed
Also published asCA1291574C, EP0304302A2, EP0304302A3
Publication number07233601, 233601, US 4991087 A, US 4991087A, US-A-4991087, US4991087 A, US4991087A
InventorsForbes J. Burkowski, Marke S. Krebs
Original AssigneeBurkowski Forbes J, Krebs Marke S
Export CitationBiBTeX, EndNote, RefMan
External Links: USPTO, USPTO Assignment, Espacenet
Computer system
US 4991087 A
Abstract
A method of operating a computer system to store and retrieve information in a database uses a signature file of the database that is divided into subsets. A word signature is mapped to a particular subset during creation of the file and the same mapping information is used to retrieve the information in response to a query word. Each word signature is a logical word signature and has two components a physical word signature and a subset designation field. In this way, when information is retrieved from the database, only that subset containing the relevant word signature is scanned. The signature file is automatically created by the system as the database is stored on the data storage modules. During retrieval, the control reviews information received from the data storage means and if a match occurs between a physical word signature for a query word and a particular physical word signature arriving from the input section, the control sends the physical word signature to the FIFO buffer in memory together with the document identifier located subsequent to the matched physical word signature. The control then moves on to process the next physical word signature received from the data storage means. If there is no match, the control ignores the physical word signature and moves on to process the next physical word signature received from the data storage means. The control is effectively capable of processing several query words in parallel.
Images(4)
Previous page
Next page
Claims(14)
We claim:
1. A method of operating a computer system to store and retrieve information in a database, said computer system having data processing means, memory means and data storage means capable of containing data records, said method comprising the steps of:
(a) storing the database on the data storage means;
(b) during loading, said system automatically creating for the database a signature file which is divided into subsets, mapping a word signature to a particular subset during creation of the file and storing said signature file subsets on said data storage means;
(c) during retrieving, after the signature file is created, entering at least one query word into the system;
(d) said system automatically creating a signature for each query word entered into the system;
(e) scanning for a word signature and retrieving the corresponding data from said database in response to a query word by using the same mapping information that was used to store the word signature in a particular subset, said system matching the signature of a query word with at least one signature in one subset of said signature file if such an appropriate word signature exists in this subset.
2. A method as claimed in claim 1 comprising the steps of, when matching a word signature in rsponse to a query word, having the system scan only that subset of the signature file containing a physical word signature that matches a signature for the query word.
3. A method as claimed in claim 2 comprising the steps of, during creation of the signature file for a particular document, automatically ignoring all common words in said document, automatically computing a logical word signature for each remaining word in the document and, automatically eliminating any duplicate logical word signatures in the document.
4. A method as claimed in claim 3 including the steps of, during the creation of the signature file, automatically dividing the signature file into a large number of subsets, generating each logical word signature so that it has two components, a physical word signature concatenated to a subset designation field, mapping all logical word signatures having the same subset designation field to the same subset, resulting in the storage in that subset of the physical word signature component of that logical word signature.
5. A method as claimed in claim 4 comprising the steps of, during the creation of the subsets, adding a document identifier, serving to identify the document from which a particular physical word signature or group of physical word signatures is derived, to that physical word signature or group of physical word signatures derived from the same document.
6. A method as claimed in claim 5 wherein the logical word signature is m-bits in length, where m ranges from 8 to 32, and has an n-bit physical word signature concatenated to a subset designation field which selects one of several subsets, where n ranges from 8 to 20.
7. A method as claimed in claim 6 wherein the representation of the document identifiers follows one or more physical word signatures stored within the signature file.
8. A method as claimed in claim 3 wherein the data storage device is a disk, said method comprising the steps of fetching one or more subsets using the data processing means in conjunction with a processor that scans a data stream emanating from said disk looking for pre-specified physical word signatures.
9. A method as claimed in claim 8 wherein each n-bit physical word signature is used as an address for a RAM table with two to the power n locations, where n is an integer ranging from 8 to 20, each location containing one bit, the processor controlled in such a fashion that it ignores physical word signatures when the addressed RAM bit location indicates a word signature of no interest to the query, and places any physical word signature into a FIFO buffer, for future reference, when the addressed RAM bit location contains the complementary bit value, the processor filtering out the physical word signatures of interest.
10. A method as claimed in claim 9 including the steps of, during creation of the signature file, generating a series of groups to make up each subset, each group having a series of physical word signatures, each group being terminated by a representation of a document identifier, each document identifier having a high, middle and low order field, the representation of the document identifier of a first group in the subset actually having a high, middle and low order field, the representations of document identifiers from subsequent groups always having a low order field but only having a middle order field or a high order field in addition to the low order field when necessary to reflect a change or changes in these fields from the document identifier represented in the immediately prior group, the document identifiers being arranged in ascending order.
11. A method as claimed in claim 10 including the steps of, during retrieval of information from the data storage means, using a processor to scan a data stream from the data storage means, the processor pushing all representations of document identifiers having a high order field into the FIFO buffer, the processor saving the last middle order field encountered in the data stream in a register for future use, said last middle order field encountered being inserted into the FIFO buffer along with the low order field that terminates a particular group only if the RAM table has indicated a match between the physical word signature derived from the query keyword and a physical word signature in the particular group.
12. A method as claimed in claim 1 wherein subsets are arranged on the data storage device with free space between consecutive subsets to allow for an increase in size of the subsets as the database is increased in size, said method comprising the steps of increasing the size of the database by adding additional information to the end thereof and in the size of the signature file by adding additional word signature groups at the end of one or more subsets.
13. A method as claimed in claim 12 wherein a logical word signature is to be associated with its corresponding word, the association between the logical word signature and the corresponding word being retained in a word dictionary, the method including the steps of choosing the subset designation field in such a way as to help in maintaining uniform lengths of all subsets and choosing the physical word signatures to ensure a one to one mapping between the logical word signatures and the corresponding word.
14. A method of operating a computer system for storing and retrieving information in a database, said computer system having a processor, a memory with address lines and a FIFO buffer, said processor having an input section, an output section and a control, said method comprising the steps of:
(i) storing the database in the data storage means;
(ii) during loading of the database, said system automatically creating for the database a signature file which is divided into subsets, mapping a word signature to a particular subset during creation of the file and storing said signature file subsets on said data storage means;
(iii) during retrieving, after the signature file is created, entering at least one query word into the system;
(iv) said system automatically creating a signature for each query word entered into the system;
(v) in response to a query word, said input section receiving information from said data storage means, said control reviewing said information and sending all word signatures to said address lines of said memory, said memory providing information to the control to enable it to determine whether a particular word signature in the input section matches a signature for said query word if a match occurs,
(a) the control sending the word signature to said FIFO buffer, the control remembering that a match has occurred and sending the document identifier located subsequent to the matched word signature to the FIFO buffer, the control then moving on to process the next word signature received from the data storage means and so on;
if there is no match,
(b) the control ignoring the word signature and moving on to process the next signature word received from the data storage means and so on;
the control thereby effectively being capable of processing several query words in parallel.
Description
BACKGROUND OF THE INVENTION

1. Field of the Invention

This invention relates to a data retrieval system and, in particular, to a method of storing data on and retrieving data from a database using a signature file that is divided into subsets, the signature file being automatically created as the data is stored.

2. Description of the Prior Art

Data storage and retrieval systems are known. With the creation of larger and larger databases and the increased use thereof, it is becoming increasingly important to have an accurate method of storing and retrieving the data in a minimum amount of time. It is also important to be able to add information to an existing database without making extensive amendments to the existing information.

In many known databases, the information is highly organized as a carefully indexed structure resident on a storage medium such as a disk. When a portion of the data is to be retrieved, the system makes use of this indexing information in order to locate the required data which may be embedded in a huge collection of similar data items. Indexing information is stored in an extra file. There are various ways of establishing said index. One possible approach is the use of a signature file, the other is the use of an inverted file. The latter approach is often used since it provides fast retrieval but it has two significant shortcomings:

(a) the size of an inverted file is extremely large, being 20% to 100% of the size of the text file itself;

(b) when new data is to be entered into the database, the inverted file must be changed and the highly structured nature of the file makes modification of said file a very time consuming process since significant portions of the file must be altered.

While signature files provide extremely fast update capabilities, they are usually not the index strategy of choice since the retrieval time can be very slow. The slow retrieval time is caused by the scanning of the entire signature file and this is a very time consuming endeavour because the required transfer time from the disk will usually be of a very lengthy duration. The invention to be described uses signature files but with a strategy that make them very competitive with the inverted file approach. Rather than scanning the entire file, a subset of it is scanned instead. This will significantly lower the access time and if the system is carefully designed the modification time is still kept fairly low.

Of course, data can always be retrieved by scanning the entire database (avoiding the use of index files) but that is extremely time consuming and therefore prohibitively expensive.

When the data storage device is a write once optical disk, an update problem occurs with this technology as it is impossible to write information onto a particular area of the optical disk more than once. Therefore, new information that would normally fit adjacent to existing information cannot be placed near that information as there is no space for it at the desired location on the optical disk. Consequently, an index structure such as an inverted file cannot be changed due to this non-erasable attribute of the storage medium. It is possible to simply create a new file in a new area of the disk but this strategy is very inefficient due to the extreme waste of disk space.

When signature files are used, new information added to the database causes the signature file to increase in size since new entries are appended to the file. Since the existing information in the signature file is not changed the use of these files is of considerable advantage in the optical disk environment.

In many cases, the system responds to a user query by retrieving one or more documents that contain one or more words that the user has specified within that query. To accomplish this, indexing facilities are used to specify the locations of required information in the database. By issuing a list of document identifiers which serve to locate the documents that contain these keywords, the index facilities working in conjunction with the query resolution software determines a final list of documents satisfying the needs of the query.

The use of signature files to locate data in a database is a known strategy. A signature file is a condensation of the information in the database. This is accomplished by representing each distinct word in a document of the database with a word signature. When the system is presented with a particular query word, it will derive the word signature that has been associated with that database word. These types of systems then cause the entire signature file to be searched using a serial scan strategy and, subsequently, based on the results of that search, all documents in the database containing the word can be found. This occurs because any word signature in the signature file is followed by a document identifier for the document that contains the word from which the word signature was derived. Consequently, when a word signature in the signature file matches the word signature derived from the query word, during the scan process, the system will capture the accompanying document identifier in order to retain the identity of documents that are pertinent to the query. These systems can still be very time consuming if the entire signature file is searched for each query word.

Optical disks are the most economical means to store databases. However, seek times for an optical disk are typically four to more than thirty times as long as the seek time for a magnetic hard disk. When databases are searched using the inverted file approach, the system may undertake several probes of the index structure with each probe possibly requiring a disk seek or disk arm movement. When a more expensive magnetic hard disk is used, the time requirements are tolerable but can become extremely undesirable when using an optical disk.

SUMMARY OF THE INVENTION

It is an object of the present invention to provide a relatively fast method of data storage and retrieval whereby only a single probe into the signature file is necessary and only part of the signature file is searched in response to a particular query keyword.

In accordance with the present invention, a method of operating a computer system to store and retrieve information in a database wherein the computer system has data processing capability, memory and data storage modules capable of containing data records is as follows:

(a) storing the database on the data storage modules;

(b) during loading, said system automatically creating for the database a signature file which is divided into subsets, mapping a word signature to a particular subset during creation of the file and storing said signature file subsets on said data storage modules;

(c) during retrieval, after the signature file is created, entering at least one query word into the system;

(d) said system automatically creating a signature for each query word entered into the system;

(e) scanning for a word signature and retrieving the corresponding data from said database in response to a query word by using the same mapping information that was used to store the word signature in a particular subset, said system matching the signature of a query word with at least one word signature in one subset of said signature word if such an appropriate word signature exists in this subset.

BRIEF DESCRIPTION OF THE DRAWINGS

In the drawings:

FIG. 1 is a diagram of a prior art hardware equipment assemblage with which the present invention can be used;

FIG. 2 is a drawing of an index structure showing three space extents after each subset;

FIG. 3 is a drawing showing that a subset is comprised of many signature groups;

FIG. 4 is a drawing of a signature group having a document identifier following each word signature;

FIG. 5 is a computer programmable process flow diagram illustrating a loading method using the present invention; and

FIG. 6 is a computer programmable process flow diagram illustrating a retrieval method using the present invention.

DESCRIPTION OF A PREFERRED EMBODIMENT

From FIG. 1, it can be seen that a data storage and retrieval system in accordance with the present invention is designed so that it can be accessed by a number of users simultaneously through user terminals 2, each user searching for different data and using different query words. The system also has a plurality of or data storage modules 4 on which the database is stored and from which the data can be retrieved.

Once the word signatures of the query words are detected in the signature file of the database, the identifiers of all documents containing the corresponding words are collected as a list of candidate documents. If desired, the software can be designed to process this list to determine whether or not a particular document satisfies additional requirements of a query. The user is then informed of the presence of the documents and may retrieve the actual documents for examination if desired.

A signature file is automatically created from the database, and is subsequently stored on the data storage device or optical disk. The signature file contains a sequence of integers (bit strings of fixed length), each integer representing the physical word signature of a significant word contained in the main text of the database. When the signature file for a particular document is automatically created, two steps are followed:

1. Common words are removed using a list of stop words;

2. A logical word signature is computed for each distinct remaining word in the document. This computation maps words (character strings) onto integer values that are m bits long, where m is an integer ranging from 8 to 32. Preferably, each logical word signature is generated so that it has two components, an n-bit physical word signature concatenated to a subset designation field, which selects one of several subsets, where n ranges from 8 to 20;

Stop words are words that generally do not contribute to the processes that serve to discriminate one document from another when the requirements of the query are to be met. They are often connecting words or common articles. For example, common words are "a", "the", "when", "where", "henceforth", etc. A list of stop words is loaded onto the system and the system automatically overlooks the stop words and does not create a word signature for them.

As shown in FIG. 2, each signature file subset, when automatically created during a database load, is a series of document signature groups appearing in the same order as the corresponding text documents (see FIG. 4). Each document signature group is comprised of a series of word signatures derived from the words in the same corresponding text document. The final entry in the signature group for one document is a representation of the identifier of that document (see FIG. 4). When a user query is to be satisfied, query words from the user query are converted into logical signatures and the signature file is scanned for matching physical word signatures. Whenever a match occurs, a representation of a document identifier at the end of the group is extracted so that the corresponding text document can be eventually located. Of course, if an appropriate word signature does not exist in any subset, then no match will occur.

It is possible to generate logical word signatures by using a hash function. This has the advantage of a fast generation and very little memory usage. However there is a drawback: since the hash encoding for a particular word signature is not guaranteed to be a one to one mapping, it is possible that different words map to the same logical word signature. During a scan operation, this multiple map can produce a false drop and an unrelated document of no interest to the user may be retrieved. This can be detected and rectified by having software check documents before they are passed to the user. False drops can be made to occur very infrequently if word signatures are long enough. The signature file can be created by using a hash function that maps words onto integer values formed from a number of bits, and creating representations of the document identifiers that are used to derive pointers to the documents.

Word signatures are usually significantly shorter than the individual text words that they represent. Also, as common words and duplicate signatures are eliminated, the signature file is much smaller than the database from which it is derived. Typically, the length of the signature file is roughly 5% to 60% of the length of the database. If the signature file contains a provision that produces related terms or synonyms that are placed in an abstract prior to the start of a document, the signature file will be somewhat enlarged.

If additional documents are appended to the database, the signature file is similarly extended by appending to the signature file additional word signatures in the form of document signature groups. The fact that the signature file can be easily updated is a distinct advantage to be gained by the use of signature files.

While the scanning of a signature file is much faster than the scanning of the entire database, even the scanning of the signature file can be fairly time consuming, particularly when the data storage device is an optical disk. For example, if a database has a size of 700 megabytes, the signature file will have a size of approximately 35 megabytes. The data streams into the data processing means from the disk at a typical rate of one megabyte per second so any scan of the signature file will take at least 35 seconds.

In order to minimize the scan time for an optical disk while retaining the advantage of an easy update, the signature file is subdivided into a reasonably large number of subsets, (for example, 256 or more subsets). During the creation of the signature file, a word signature is mapped to a particular subset and the same mapping information is used when the subset is scanned for the word signature in response to a query word.

Various word signature formats are possible. The following format will serve as an example: 15 bit physical word signatures are stored in two consecutive bytes on a disk with the most significant bit used as a steering bit, one setting of this bit will signify that this 16-bit word contains a physical word signature while the complementary setting will signify that this 16-bit word serves in the indentification of the document.

Scan time will be considerably reduced since, in response to a query word, it is only necessary to scan a small fraction of the total signature file. Subdivision of the signature file will lead to more total space consumption for the file when compared to signature files that are not subdivided. Since the group of all word signatures for a document is now dispersed across several subsets, a document identifier is needed in every subset that contains a portion of that total group.

If we assume that each two byte word signature is followed by a two byte document pointer, then a word signature, in this worst case, will occupy four bytes. When the signature file is not subdivided, each entry is typically a 24 bit signature that occupies three bytes, excluding document numbers. Thus, the total size of a signature file that has been divided into subsets is no more than one-third larger than a non-subdivided signature file.

If a 700 megabytes, using the example described earlier, the size of the signature file after being subdivided will be an additional 33% or 99.3 megabytes in all. When this subdivided file is spread over 256 subsets, each subset will have an average size of approximately 0.36 megabytes. With a transfer rate of one megabyte per second, an average subset can be transferred in 0.36 seconds.

A signature subset is scanned most effectively if no disk seek is necessary during that scan. This requires that a subset be maintained in contiguous sectors of the disk even though it will grow with the addition of new information. This is done by providing free space between the consecutive subsets as shown in FIG. 2. This free space will be gradually utilized as the database increases in size. When the data storage module is a disk, the data processing means is used to fetch one or more subsets from the disk. The processor looks for pre-specified physical word signatures.

The system helps to ensure the predictable growth of subsets by exercising some care when it decides which subset will hold a particular word signature.

When a word is first encountered in the database it is assigned to the subset which currently has the shortest length. For each subset, the processor keeps track of the highest physical word signature previously assigned. The new entry is given a word signature equal to this maximum plus one. Once the subset identifier is defined and the physical word signature is also defined, the processor will concatentate the two to form the logical word signature which can then be established in the database word dictionary. This involves a two pass scan of the text in the database. During the first pass, the system creates a list of all distinct words in the database keeping track of the number of documents that contain each word. This document count will serve to properly estimate, for each word, the total storage requirements of its word signature groups within a subset. The words are then sorted in ascending order of document counts. It is then a simple matter to assign a subset to the word signature of each word in such a way that the length of the signature subsets are as uniform as possible. This assignation is saved in the word dictionary so that during the second pass through the database, the system will compose logical signatures by concatenating the subset assignment to the physical word signature. When the association between the logical word signature and the corresponding word is retained in a keyword dictionary, the values of the subset designation fields within their logical word signatures are chosen to help in maintaining the uniform lengths of all subsets. The physical word signatures are chosen to ensure a one to one mapping between the logical word signatures and the corresponding word. The physical word signature for a word can be uniquely defined once the system knows which words are associated with a particular subset. This is done by simply numbering the distinct keywords assigned to the subset with consecutive integers. These numbers are taken to be the physical word signatures.

Depending on the application, the system could respond to the user in a variety of ways. When a particular query words is fed into the system, it is expected that the system will search only that subset containing the physical word signature matching the physical word signature of the query word. The subset will contain the document identifiers of all documents in the database that contain that query word. Using the document identifiers the documents can then be retrieved from the database without scanning the entire database.

The number of bits used in a physical word signature can range over a number of possible values depending on the requirements of the system. An n-bit physical word signature is used as an address for a RAM table with two to the power n locations, where n is an integer ranging from 8 to 20. Each location contains one bit and the subset scanning is controlled in such a fashion that it ignores word signatures when the RAM indicates a word signature is of no interest to the query. Otherwise, it places the physical word signature into the FIFO buffer.

In the following embodiment we will assume a length of 15 bits for the physical word signatures. Each 15 bit physical word signature is used as an address for a 32K by one RAM table which has contents previously defined by software running on the central processing unit (henceforth CPU). If the bit location accessed contains a binary value which indicates that this physical word signature is not being sought by this particular query, the physical word signature is ignored. If the bit location accessed contains the complementary value, then the physical word signature and the following document identifier are placed in a first in first out (FIFO) buffer for future reference by the CPU. When a particular physical word signature is ignored, the system proceeds to examine the next physical word signature. The preferred number of bits used in the logical word signature is m-bits in length, where m ranges from 8 to 32.

In the interest of minimizing the disk space utilized by the signature file, the representations of document identifiers should occupy, on average, as little space as possible. One manner in which this can be done, during creation of the signature file, is by treating the document number as a value that is made up of three fields, a high order field labelled H, a middle order field labelled M and a low order field labelled L as follows: ##STR1##

Although the document numbers are not necessarily consecutive, they will occur in ascending order within a particular subset. Therefore, the H or M fields are placed into the next representation of the document identifier only when a lower order field goes through a zero value since the last representation. During a scan operation, the H field is always pushed into the FIFO buffer whenever it is encountered. When an M field appears in the stream, it is saved for future use in a register that will be called the MREG register. The group of word signatures for each document can then be terminated with a single byte containing the L field. The M field is inserted prior to an L field only if the M value has increased beyond the value associated with the previous L field. When a representation of a document identifier is to be entered into the FIFO buffer, it is combined with the last seen M field (the one that was saved in MREG). The software processing the FIFO buffer output will combine M, L pairs with the H field last seen in the stream. In this fashion, references to a document can be handled with a small space overhead that is slightly more than one byte per document group on average. While the representation of the document identifier might be an `L field` only or an `M and L field` only, the actual document identifier always contains all three fields.

Table 1 presents an example of encodings for bytes and words in a signature file subset. In Table 1, LSSA and HSSA represent low order and high order portions of the signature sector address, which is included at the beginning of each sector as a redundent check on the source of the word signature stream.

In more general terms, during creation of the signature file, a series of groups are generated to make up each subset. Each group has a series of physical word signatures and is terminated by a representation of a document identifier. The actual document identifier has a high, middle and low order field. The representation of the document identifier of the first group in the subset actually has a high, middle and low order field. The representations of document identifiers from subsequent groups always have a low order field but only have a middle order field or a high order field when necessary to reflect a change or changes in these fields from the document identifier represented in the immediately prior group. The document identifiers are arranged in ascending order. Thus, while the representation of the document identifier for a particular group might have a low order field only, the actual document identifier, for that group, will have a high, middle and low order field. As a variation, if there was no concern with saving space, all representations of document identifiers could actually have three fields.

                                  TABLE 1__________________________________________________________________________Type of Subset          Bit LengthEntry     Format        of Entry                         Action to be Taken__________________________________________________________________________Word Signature      ##STR2##     16    Push into FIFO if a match is detectedDocument Number L Field      ##STR3##      8    Combine with MREG and push into FIFO if a                         signature in this group is detectedDocument Number M Field      ##STR4##      8    Save in MREGDocument Number H Field      ##STR5##     16    Push into FIFOLow Order Portion of Signature Sector Address      ##STR6##     16    Push into FIFOHigh Order Portion of Signature  Sector Address      ##STR7##     16    Push into FIFO__________________________________________________________________________

During retrieval of information from the data storage areas, the processor pushes all document identifiers having a high order field into the FIFO buffer. The processor saves the last middle order field encountered in the data stream in a register for future use. The last middle order field encountered is inserted into the FIFO along with the low order field that terminates a particular group only if the RAM has indicated a match between the physical word signature of the query word and a physical word signature in that particular group.

In FIG. 5, there is shown a process flow diagram for loading the database. During the loading of the database, the processor automatically creates a word signature file with document identifiers, the identifiers being used for retrieving a document from the database in response to a query word. During loading, the processor, extracts each word of text and determines whether the word is a common word or not. If the word is a common word, the processor passes the word by. If the word is not a common word, the processor next determines whether the word is in the word dictionary. If the word is not in the word dictionary, the processor finds an identifier for the shortest subset. Next, the processor generates a physical word signature for the word and concatenates the identifier for the subset and the physical word signature to form a logical word signature. Then, the logical word signature is placed in the word dictionary.

If the word is already in the word dictionary, the physical word signature and a representation of the document identifier are placed into the subset designated by the subset identifier by overwriting the previous document identifier if it is the same. Next, it is determined whether the subset is filled. If not, the next word of text of the database is extracted and the process is repeated. If the subset is filled, the subsets are automatically shifted to leave free space for each subset in proportion to the current size of that subset. After the shifting occurs, the next word of the text of the database is extracted and the process is repeated.

In FIG. 6, there is shown a process flow diagram for the retrieval method. During retrieval, the processor first consults the word dictionary to find the subset identifier and physical word signature of the given query word. The processor then starts to scan the designated subset. From the first subset entry, the processor compares the subset entry to the physical word signature of the query word to determine whether or not there is a match. If there is no match, the processor moves onto the next subset entry and repeats the comparison. If there is a match, the processor picks up the representation of the document identifier. If the identifier contains an H field, this is retained for future use. If the identifier contains only the M and L fields, the full representation of the document identifier is generated by concatenating the previously saved H field to the M and L fields. Whenever an M field is part of the representation, it is saved for future use in a register called MREG. If the document identifier encountered during retrieval contains only the L field, it is combined with both of the previously saved H and M fields to obtain the full representation of the identifier. The processor then moves onto the next subset entry and the process is repeated for the entire subset.

The advantage of the method of the present invention over other existing methods such as inverted lists arises from the fast update (database append) activity that it supports. When new information is added to the database index the subsets expand into available space that is provided for this purpose. Since words share subsets the growth of a subset is dependent on the average frequency distribution of its constituent words. This property makes the growth of a subset very predictable. The method reassigns free space, when necessary to the subsets in proportion to their current size. Studies have shown that this predictable expansion promotes a uniform consumption of the free space across almost all subsets during load operations. This improves load performance because the processor does not have to spend time frequently shifting the subsets to maintain their contiguity on the disk due to any subset overflow. The method anticipates that occasionally free space is entirely used up and a subset operation is necessary. The significant improvement of this indexing technique rests on the fact that such a shift occurs only rarely and so load operations are fast since they are mainly the simple insertion of new index information.

Numerous variations, within the scope of the attached claims, will be readily apparent to those skilled in the art.

Patent Citations
Cited PatentFiling datePublication dateApplicantTitle
US4068301 *Oct 14, 1975Jan 10, 1978Ricoh Company, Ltd.Data storage device comprising search means
US4085446 *Aug 17, 1976Apr 18, 1978Fumitaka NagamuraData storage and retrieval system
US4255796 *Feb 14, 1978Mar 10, 1981Bell Telephone Laboratories, IncorporatedAssociative information retrieval continuously guided by search status feedback
US4318184 *Sep 5, 1978Mar 2, 1982Millett Ronald PInformation storage and retrieval system and method
US4358824 *Dec 28, 1979Nov 9, 1982International Business Machines CorporationOffice correspondence storage and retrieval system
US4554631 *Jul 13, 1983Nov 19, 1985At&T Bell LaboratoriesKeyword search automatic limiting method
Referenced by
Citing PatentFiling datePublication dateApplicantTitle
US5148541 *Nov 3, 1989Sep 15, 1992Northern Telecom LimitedMultilingual database system including sorting data using a master universal sort order for all languages
US5265065 *Oct 8, 1991Nov 23, 1993West Publishing CompanyMethod and apparatus for information retrieval from a database by replacing domain specific stemmed phases in a natural language to create a search query
US5307266 *Aug 7, 1991Apr 26, 1994Hitachi, Ltd.Information processing system and method for processing document by using structured keywords
US5319745 *Sep 16, 1992Jun 7, 1994Societe Nationale Industrielle Et AerospatialeMethod and apparatus for processing alphanumeric and graphic information to create a data base
US5319779 *Feb 19, 1992Jun 7, 1994International Business Machines CorporationSystem for searching information using combinatorial signature derived from bits sets of a base signature
US5355476 *Dec 2, 1991Oct 11, 1994Casio Computer Co., Ltd.File update apparatus for generating a matrix representing a subset of files and the update correspondence between directories and files
US5367672 *Aug 2, 1991Nov 22, 1994Kabushiki Kaisha ToshibaApparatus for retrieving data using the remainder of a hashed search key to identify a group of keys and comparing the key to the group of keys
US5396624 *Aug 3, 1993Mar 7, 1995Visa International Service AssociationAccount file for off-line transaction authorization
US5418948 *Sep 8, 1993May 23, 1995West Publishing CompanyConcept matching of natural language queries with a database of document concepts
US5423040 *Jan 14, 1994Jun 6, 1995International Business Machines CorporationComputer-implemented system
US5535382 *Nov 17, 1993Jul 9, 1996Ricoh Company, Ltd.Document retrieval system involving ranking of documents in accordance with a degree to which the documents fulfill a retrieval condition corresponding to a user entry
US5542090 *Jul 27, 1994Jul 30, 1996Xerox CorporationText retrieval method and system using signature of nearby words
US5586218 *Aug 24, 1995Dec 17, 1996Inference CorporationAutonomous learning and reasoning agent
US5692176 *Nov 22, 1993Nov 25, 1997Reed Elsevier Inc.Associative text search and retrieval system
US5696963 *Apr 14, 1995Dec 9, 1997Waverley Holdings, Inc.System, method and computer program product for searching through an individual document and a group of documents
US5708829 *Sep 25, 1996Jan 13, 1998Wang Laboratories, Inc.For generating an index for a collection of words
US5724576 *Apr 3, 1995Mar 3, 1998Prime Arithmetics, Inc.Method and apparatus for the generation and manipulation of data structures
US5742806 *Jan 31, 1994Apr 21, 1998Sun Microsystems, Inc.Apparatus and method for decomposing database queries for database management system including multiprocessor digital data processing system
US5745900 *Aug 9, 1996Apr 28, 1998Digital Equipment CorporationMethod for indexing duplicate database records using a full-record fingerprint
US5748953 *May 18, 1995May 5, 1998Hitachi, Ltd.Document search method wherein stored documents and search queries comprise segmented text data of spaced, nonconsecutive text elements and words segmented by predetermined symbols
US5758152 *Dec 6, 1990May 26, 1998Prime Arithmetics, Inc.Method and apparatus for the generation and manipulation of data structures
US5761497 *Jun 7, 1995Jun 2, 1998Reed Elsevier, Inc.Associative text search and retrieval system that calculates ranking scores and window scores
US5771378 *Jun 7, 1995Jun 23, 1998Reed Elsevier, Inc.Associative text search and retrieval system having a table indicating word position in phrases
US5848409 *Aug 4, 1997Dec 8, 1998Smartpatents, Inc.System, method and computer program product for maintaining group hits tables and document index tables for the purpose of searching through individual documents and groups of documents
US5884303 *Feb 6, 1997Mar 16, 1999International Computers LimitedParallel searching technique
US5913209 *Jul 7, 1997Jun 15, 1999Novell, Inc.Full text index reference compression
US5950214 *Apr 10, 1998Sep 7, 1999Aurigin Systems, Inc.System, method, and computer program product for accessing a note database having subnote information for the purpose of manipulating subnotes linked to portions of documents
US5963942 *Dec 24, 1996Oct 5, 1999Fujitsu LimitedPattern search apparatus and method
US5966703 *Apr 3, 1998Oct 12, 1999Digital Equipment CorporationTechnique for indexing information stored as a plurality of records
US5970497 *Apr 27, 1998Oct 19, 1999Digital Equipment CorporationMethod for indexing duplicate records of information of a database
US5991751 *Jun 2, 1997Nov 23, 1999Smartpatents, Inc.System, method, and computer program product for patent-centric and group-oriented data processing
US5991780 *Apr 3, 1998Nov 23, 1999Aurigin Systems, Inc.Computer based system, method, and computer program product for selectively displaying patent text and images
US6018749 *Apr 9, 1998Jan 25, 2000Aurigin Systems, Inc.System, method, and computer program product for generating documents using pagination information
US6119114 *Sep 17, 1997Sep 12, 2000Smadja; FrankMethod and apparatus for dynamic relevance ranking
US6173298Sep 17, 1997Jan 9, 2001Asap, Ltd.Method and apparatus for implementing a dynamic collocation dictionary
US6185559May 9, 1997Feb 6, 2001Hitachi America, Ltd.Method and apparatus for dynamically counting large itemsets
US6216123 *Jun 24, 1998Apr 10, 2001Novell, Inc.Method and system for rapid retrieval in a full text indexing system
US6219818Feb 18, 1999Apr 17, 2001Netmind Technologies, Inc.Checksum-comparing change-detection tool indicating degree and location of change of internet documents
US6339767 *Aug 29, 1997Jan 15, 2002Aurigin Systems, Inc.Using hyperbolic trees to visualize data generated by patent-centric and group-oriented data processing
US6374250 *Feb 3, 1997Apr 16, 2002International Business Machines CorporationSystem and method for differential compression of data from a plurality of binary sources
US6389434Apr 9, 1998May 14, 2002Aurigin Systems, Inc.System, method, and computer program product for creating subnotes linked to portions of data objects after entering an annotation mode
US6499026Sep 15, 2000Dec 24, 2002Aurigin Systems, Inc.Using hyperbolic trees to visualize data generated by patent-centric and group-oriented data processing
US6687687 *Jul 26, 2000Feb 3, 2004Zix Scm, Inc.Dynamic indexing information retrieval or filtering system
US6728659Jul 14, 1998Apr 27, 2004Tlv Company, LimitedEquipment inspection and evaluation system, equipment management system, and computer-readable record medium with equipment management program stored therein
US6772093Jan 24, 2001Aug 3, 2004Tlv Company, LimitedEquipment inspection and evaluation system, equipment management system, and computer-readable record medium with equipment management program stored therein
US6850848 *Jan 24, 2001Feb 1, 2005Tlv Company, LimitedEquipment inspection and evaluation system, equipment management system, and computer-readable record medium with equipment management program stored therein
US6877137Dec 7, 1999Apr 5, 2005Rose Blush Software LlcSystem, method and computer program product for mediating notes and note sub-notes linked or otherwise associated with stored or networked web pages
US6880128Jan 24, 2001Apr 12, 2005Tlv Company, LimitedEquipment inspection and evaluation system, equipment management system, and computer-readable record medium with equipment management program stored therein
US6912655Aug 9, 1999Jun 28, 2005Tristrata Security Inc.Network security architecture system utilizing seals
US6917709 *Dec 16, 2002Jul 12, 2005Parascript LlcAutomated search on cursive records not having an ASCII index
US7146376 *Apr 24, 2003Dec 5, 2006International Business Machines CorporationData abstraction model driven physical layout
US7152074 *Sep 19, 2003Dec 19, 2006International Business Machines CorporationExtensible framework supporting deposit of heterogenous data sources into a target data repository
US7158983Sep 23, 2002Jan 2, 2007Battelle Memorial InstituteText analysis technique
US7257706May 6, 2005Aug 14, 2007Tristrata Security, Inc.Method of securing a document in a system and controlling access to the document and a seal for use in the method
US7487149 *Mar 10, 2004Feb 3, 2009Oracle International CorporationMethod and apparatus for providing query-level security in a database system
US7487542Jan 14, 2004Feb 3, 2009International Business Machines CorporationIntrusion detection using a network processor and a parallel pattern detection engine
US7523126Jun 22, 2002Apr 21, 2009Rose Blush Software LlcUsing hyperbolic trees to visualize data generated by patent-centric and group-oriented data processing
US7620632Dec 7, 2004Nov 17, 2009Skyler Technology, Inc.Method and/or system for performing tree matching
US7627591Dec 7, 2004Dec 1, 2009Skyler Technology, Inc.Method and/or system for manipulating tree expressions
US7630995Dec 7, 2004Dec 8, 2009Skyler Technology, Inc.Method and/or system for transmitting and/or receiving data
US7636727Dec 6, 2004Dec 22, 2009Skyler Technology, Inc.Enumeration of trees from finite number of nodes
US7676453 *Apr 22, 2004Mar 9, 2010Oracle International CorporationPartial query caching
US7681177Feb 23, 2006Mar 16, 2010Skyler Technology, Inc.Method and/or system for transforming between trees and strings
US7716060Feb 23, 2001May 11, 2010Germeraad Paul BPatent-related tools and methodology for use in the merger and acquisition process
US7734780Mar 17, 2008Jun 8, 2010Microsoft CorporationAutomated response to computer users context
US7739255Sep 1, 2006Jun 15, 2010Ma Capital LllpSystem for and method of visual representation and review of media files
US7743249Feb 14, 2007Jun 22, 2010Tristrata Security, Inc.Method of securing a document in a system and controlling access to the document and a seal for use in the method
US7779015Nov 8, 2004Aug 17, 2010Microsoft CorporationLogging and analyzing context attributes
US7797336May 4, 2001Sep 14, 2010Tim W BlairSystem, method, and computer program product for knowledge management
US7801923Dec 6, 2004Sep 21, 2010Robert T. and Virginia T. Jenkins as Trustees of the Jenkins Family TrustMethod and/or system for tagging trees
US7877686Oct 15, 2001Jan 25, 2011Microsoft CorporationDynamically displaying current status of tasks
US7882147Dec 6, 2004Feb 1, 2011Robert T. and Virginia T. JenkinsFile location naming hierarchy
US7899821Apr 26, 2006Mar 1, 2011Karl SchiffmannManipulation and/or analysis of hierarchical data
US7945859Dec 17, 2008May 17, 2011Microsoft CorporationInterface for exchanging context data
US7949581Sep 7, 2006May 24, 2011Patentratings, LlcMethod of determining an obsolescence rate of a technology
US7949728Aug 31, 2006May 24, 2011Rose Blush Software LlcSystem, method, and computer program product for managing and analyzing intellectual property (IP) related transactions
US7962511Apr 29, 2003Jun 14, 2011Patentratings, LlcMethod and system for rating patents and other intangible assets
US7966328Aug 31, 2006Jun 21, 2011Rose Blush Software LlcPatent-related tools and methodology for use in research and development projects
US8020104Jan 11, 2005Sep 13, 2011Microsoft CorporationContextual responses based on automated learning techniques
US8037102Dec 6, 2004Oct 11, 2011Robert T. and Virginia T. JenkinsManipulating sets of hierarchical data
US8099415 *Sep 8, 2006Jan 17, 2012Simply Hired, Inc.Method and apparatus for assessing similarity between online job listings
US8103665May 11, 2009Jan 24, 2012Microsoft CorporationSoliciting information based on a computer user's context
US8126979Apr 13, 2010Feb 28, 2012Microsoft CorporationAutomated response to computer users context
US8181113Oct 27, 2008May 15, 2012Microsoft CorporationMediating conflicts in computer users context data
US8239945Dec 14, 2008Aug 7, 2012International Business Machines CorporationIntrusion detection using a network processor and a parallel pattern detection engine
US8316059Dec 27, 2005Nov 20, 2012Robert T. and Virginia T. JenkinsEnumeration of rooted partial subtrees
US8346724Dec 8, 2008Jan 1, 2013Microsoft CorporationGenerating and supplying user context data
US8356040Mar 20, 2006Jan 15, 2013Robert T. and Virginia T. JenkinsMethod and/or system for transforming between trees and arrays
US8443339Feb 8, 2010May 14, 2013Robert T. and Virginia T. JenkinsMethod and/or system for transforming between trees and strings
US8489997May 7, 2010Jul 16, 2013Microsoft CorporationSupplying notifications related to supply and consumption of user context data
US8612461Nov 30, 2009Dec 17, 2013Robert T. and Virginia T. JenkinsEnumeration of trees from finite number of nodes
US8615530Dec 27, 2005Dec 24, 2013Robert T. and Virginia T. Jenkins as Trustees for the Jenkins Family TrustMethod and/or system for tree transformation
US8626712Jun 28, 2010Jan 7, 2014Microsoft CorporationLogging and analyzing computer user's context data
US8626777Oct 13, 2009Jan 7, 2014Robert T. JenkinsMethod and/or system for manipulating tree expressions
US8677248May 14, 2009Mar 18, 2014Microsoft CorporationRequesting computer user's context data
WO1996007148A1 *Aug 22, 1995Mar 7, 1996Andrei Valentinovic KondratievTextual or graphic data transmission system and a consumer information device for the system
Classifications
U.S. Classification1/1, 707/E17.036, 707/E17.058, 707/E17.009, 707/999.003
International ClassificationG06F17/30
Cooperative ClassificationY10S707/99933, G06F17/30017, G06F17/30949
European ClassificationG06F17/30Z1C, G06F17/30E
Legal Events
DateCodeEventDescription
Apr 18, 1995FPExpired due to failure to pay maintenance fee
Effective date: 19950208
Feb 5, 1995LAPSLapse for failure to pay maintenance fees
Sep 13, 1994REMIMaintenance fee reminder mailed